Also, the up front production costs do not need to be recaptured on first watching. The 90M to produce Marco Polo creates and asset that can be sold, traded, etc... Eventually Netflix will have so much good, custom content that a new person joining will take years to go through it all if possible. What we are watching now is them bootstrap that process, but it will not always be that way.
OTOH, it would appear in many cases those rights aren't going to Netflix directly. See
https://www.wired.com/2014/03/comcast-bites-netflix-snagging...
, for example.
(Of course, this is how Netflix is getting so many shows made - they're trading those rights to the production company in exchange for not footing the entire production cost.)
It's worth noting, though, that those will scale strongly with the initial success of the show. Breaking Bad DVD sales were huge. The sales of DVDs of, say, "Alphas" on Sci-Fi will not be nearly as impressive.
As for subscribe numbers - I don't have hard figures, but based on years of persuading people to watch video-based narrative content I'd guess 5% of total viewers for a weak show, 10% for a strong show or one that's capturing a new audience, 18% for one that's both very strong and targeted at a radically new audience. Occasional outliers like Breaking Bad and Game Of Thrones will probably provide higher capture for the subscription services showing them, but they're 1-3 times a decade phenomena.
I haven't subscribed ever for just a single show but I tend to subscribe for one month to watch a few shows and then unsubscribe for maybe 3-4 months before returning.
I have a UK account but live most of the time in Spain. The offering is similar in size. I have bothered a couple of times to go via vpn to get the US version but not found it so much better that it's worth the extra effort.
I have an amazon prime account too that I have never cancelled. Partly because I sometimes use the free delivery option when I'm in the UK but also with amazon I find new stuff worth watching tends to come along a bit more frequently.
Netflix still has shitty recommendations, and annoying pushing of content I'm not interested in. But by the time I get through all the things I actually want to watch as of today I'd be amazed if there wasn't almost as much new content. In other words I don't think I'll ever run out.
I don't know about Netflix, but anecdotally, I subscribed to HBO just for Westworld alone.
So, a failure is probably worth ~10m and a success can be worth 20+x that. Remember eventually they end up with a back catalog that's large enough to be self sustaining.
For example, a show no one ever even heard of "Tyrant", was canceled for averaging 700,000 viewers.
http://uk.businessinsider.com/netflix-marco-polo-tv-show-bud...
Let's look at a 380,000 viewer show. It's very unlikely that ever single one of those viewers paid for Netflix for a solid year just to get access to that show (which will have all its episodes released at once).
So let's assume it has significant cult appeal, meaning a full 25% of its viewers are primarily subscribed for that show and things like it. (I'd say that's likely to be very high, based on my experience in the narrative video world, but let's be optimistic.) And let's say, based on subscribing for that show and forgetting to cancel for a couple of months (or subsequently being retained by another 380k viewer show), that Netflix gets 3 months' worth of revenue from them.
So that's 380k/4 = 95,000 * $7.99 * 3 = $2.28m revenue attributable to that show.
Of the shows you mention:
Marco Polo cost $90m to produce for the entire season.
Black Mirror doesn't have figures online, but it'll be around $3m per episode if it's similar to other BBC shows. So that's $18m.
Unbreakable Kimmy Schmidt doesn't have figures available, but based on Tina Fey's fees for other shows and general knowledge of how much TV costs, I'll guess at about $3m an episode. (Might be cheaper if they're being clever about it, so could be as low as $2m) So that's $39m.
So, assuming that Netflix are just getting revenue from retained subscribers, and they get the figures above, we're looking at losses of:
$88m for Marco Polo.
$16m for Black Mirror.
$37m for Unbreakable Kimmy Schmidt.
Obviously the situation is far more complex than that - they'll be financing with coproduction deals, they're looking for growth rather than revenue, they also acquire a lot of content cheaply (very, very cheaply in some cases) through licensing - but it's certainly not rosy enough that they'll be happy with 380k viewers for shows they're making.